Re: [lwip-users] IP forwarding to/from PPP working for one netif, but not another

[address@hidden]

Am 30.01.2020 um 22:02 schrieb Sylvain Rochet:
> Hi Simon,
>
> On Thu, Jan 30, 2020 at 09:31:11PM +0100, address@hidden wrote:
> > Am 30.01.2020 um 20:42 schrieb Sylvain Rochet:
> > > On Wed, Jan 29, 2020 at 08:15:11PM -0800, Andrew Pullin wrote:
> > > > Host2:
> > > >      - Micontroller device, ESP32
> > > >      - has both WiFi and Ethernet PHY hardware (IP101GRI Ethernet PHY), used
> > > > mutually exclusively
> > > >      - Running lwIP
> > > >      - WiFi and Ethernet are independently verified to work to reach WAN from
> > > > this host
> > > >      - PPPoS server
> > > >      - ppp0 : 10.0.0.1/32
> > > >          - IP for server and client are hard-coded in
> > > >          - I believe it is /32, as the netmask is reported as 255.255.255.255
> > > >      - When running w/ WiFi:
> > > >          - wlan0 : ip 10.0.200.136/16  (IP and /16 from DHCP server)
> > > >      - When running w/ Ethernet:
> > > >          - eth0 : ip 10.0.200.113/16  (IP and /16 from DHCP server)
> >
> > One odd thing I find here is that lwIP does not support 2 interfaces in
> > the same subnet. So you would have to take care that only one netif at a
> > time has its link set up (or its address set or its administrative
> > status up, or whatever). In any case, having both wlan0 and eth0 up and
> > running with an address in the same subnet will result in packets being
> > accepted on both netifs but transmitted on whichever was registered last
> > (or was it first?).
>
> I assumed that both were not active at the same time, good catch, this
> might be an issue as well.
>
>
> > > This is not "connection sharing", this is IP routing, not NAT !,
> > > "connection sharing" is usually used for hosts that do SNAT on their
> > > Interface facing Internet.
> > >
> > > IP routing needs a complete routing table on all hosts telling how to
> > > reach (i.e. which next-hop to use) every host of the whole network.
> > >
> > > Example with 2 subnets and 3 Hosts:
> > >
> > > A-eth0  --------------------- eth0-B-eth1 --------------------- eth0-C
> > >     192.168.0.1/24    192.168.0.2/24   172.16.1.1/24     172.16.1.2/24
> > >
> > > Host A routing table:
> > > 192.168.0.0/24 dev eth0
> > > 172.16.1.0/24 via 192.168.0.2
> > >
> > > Host B routing table:
> > > 192.168.0.0/24 dev eth0
> > > 172.16.1.0/24 dev eth1
> > >
> > > Host C routing table:
> > > 192.168.0.0/24 via 172.16.1.1
> > > 172.16.1.0/24 dev eth0
> > >
> > > dev = directly connected
> > > via = next-hop, gateway if you prefer
> > >
> > > What you could see is that each host have a similar routing table, just
> > > the next-hop changes. Default route is just a way to reduce the routing
> > > table size by removing all entries that have an identical next-hop.
> >
> > Which brings me to a question: how do we define such routing rules in
> > lwIP? Hmm...
>
> I'm surprised by the question :-) In lwIP, "dev" rules don't need to be
> added, the ip4_route function "create" them dynamically using the
> interface address/netmask.

Well, yes, my bad. The question wasn't phrased well. The thing I don't
understand (because I'm not using it) probably all boils down to using
lwIP as a PPP 'proxy' to remote networks without doing NAT at the same
time (which we don't support in mainline). That's not standard routing...

> But thanks you for asking that because there is a huuuuuge issue about
> that with the aforementioned case. Since "dev" rules are not ordered
> because we just foreach on interfaces list until a match, we don't
> respect that smaller subnets should be checked and matched first. So if
> an IP subnet is within the IP subnet of another interface, although
> perfectly legal configuration but *bad* design, it does not work in
> lwIP. In lwIP it actually depends on the order interfaces were added :-)

Which results in me citing my mantra "lwIP does *not* support multiple
netifs in the same subnet". (And for those who ask why: most of the
time, this is just bad network desing. For the rest where it is
intended: we could make it work, but the burdon on the rest of us not
needing it has been too much, up to now.)

> For "via" routes, we have LWIP_HOOK_IP4_ROUTE.

Yeah, well, this is probably not used much, being a hook that people
need to implement on their own? A default implementation for a routing
algorithm woule be nice...

Regards,
Simon